The comprehensive search yielded 42 titles in Cochrane, 849 in EMBASE, and 958 in PubMed. After removing duplicates, 1,836 citations were selected as potentially relevant. The titles and abstracts were screened, and then 33 full-text articles were selected for further analysis. Twenty-nine articles were excluded, six of which were duplicated, 11 did not include ventilator duration, four did not include sepsis, three were based on the same data, two were reviews, two were retrospective studies and one did not include DEX. Finally, four randomized clinical trials with 349 patients were included in this meta-analysis [7, 8, 13, 16] (Figure 1).
Study characteristics and quality
Among the four trials, one was published in Chinese, and three were published in English. One study was double-blinded and one was blinded-endpoint, while others were not blinded [13, 16]. All patients were adults. The experimental group was DEX, and the control included propofol[8, 13, 16], lorazepam, and midazolam. The doses of DEX ranged from 0.1 μg /kg/hr to 2.5 μg /kg/hr. In one study, the patients were maintained at a Ramsay sedation score < 2. In other studies, the target sedation levels were RASS score of 1, -1 to -2, and 0 during the day and -2 during the night. The characteristics of the included studies and a summary of the durations of mechanical ventilation and/or 28-day ventilator-free days are shown in Table 1. Two studies reported the number of ventilator-free days [7, 8], but the other two did not [13, 16]. The duration of mechanical ventilator was available in three studies [8, 13, 16]. Figure 2 and Figure 3 show summaries of the risk of bias.
The quality of evidence in the included studies ranged from moderate to high (Table 2).
Three trials reported the duration of mechanical ventilation as an outcome [8, 13, 16]. Tasdogan et al. expressed the duration of mechanical ventilation in 37 survivors as a median (min–max) and in three non-survivors as the number of days requiring mechanical ventilation . Because we were unable to obtain raw data, we only pooled the data on the survivors in our meta-analysis. Among three trials, DEX was compared with propofol and propofol and midazolam . However, Guo’s study used two control groups (i.e., a propofol group and a midazolam group) , so we pooled the data accordingly. When the data were pooled in the propofol group, the fixed-effects analysis indicated that the use of DEX was not associated with a short duration of mechanical ventilation (MD 0.65, 95% CI, -0.13 to 1.42, P = 0.10; P for heterogeneity = 0.15, I2 = 47%) (n = 268) (Figure 4). When the data were pooled in the midazolam group, the meta-analysis also indicated that the sedation of DEX did not shorten the duration of mechanical ventilation (MD 0.07, 95% CI, -1.58 to 1.72, P = 0.94; P for heterogeneity = 0.08, I2 = 60%.) (n = 267) (Figure 5).
Data on 28-day ventilator-free days were available in two RCTs 16, 18, but the outcomes were opposite. Pandharipande et al. reported that the septic patients who received DEX had more ventilator-free days than those who did not receive DEX , whereas Kawazoe et al. concluded that DEX did not increase the number of ventilator-free days in septic patients. Our meta-analysis yielded a fixed-effect estimate of less ventilator-free days in patients who were not given DEX than those who were given DEX (MD 3.57, 95% CI, 0.26 to 6.89, P = 0.03; P for heterogeneity = 0.45, I2 = 0%) (n = 264) (Figure 6).
Data on 28-day mortality were available in all the RCTs included in our meta-analysis. Pandharipande et al.  reported that septic patients sedated with DEX had a lower risk of death than those who did not receive DEX, whereas Tasdogan  and Guo  reported DEX did not improve the mortality rate. Kawazoe et al.  showed that DEX resulted in an 8% reduction in 28-day mortality even though the results were not statistically significant. Our meta-analysis indicated that compared with no DEX sedation, DEX improved short-term mortality (RR 0.61, 95% CI, 0.49 to 0.94, P = 0.02; P for heterogeneity = 0.67, I2 = 0.) (n = 334) (Figure 7).
Sensitivity analysis and publication bias
The tests conducted to determine heterogeneity in the duration of mechanical ventilation revealed outliers in the results of Guo et al. (2016). We removed this study to eliminate heterogeneity, but the results were unchanged (MD 0.77, 95% CI, -0.02 to 1.56, P = 0.06; P for heterogeneity = 0.30, I2 = 7%.) (n = 238) (Figure 8). The funnel plot showed no evidence of significant publication bias in the results of the duration of ventilator-free days and mortality (Figure 9 and Figure 10).